Original llama.cpp quant methods: q4_0, q4_1, q5_0, q5_1, q8_0
I have quantized these 'original' quantisation methods using an older version of llama.cpp so that they remain compatible with llama.cpp as of May 19th, commit 2d5db48.
They should be compatible with all current UIs and libraries that use llama.cpp, such as those listed at the top of this README.
These new quantisation methods are only compatible with llama.cpp as of June 6th, commit 2d43387.
They will NOT be compatible with koboldcpp, text-generation-ui, and other UIs and libraries yet. Support is expected to come over the next few days.
Explanation of the new k-quant methods
The new methods available are:
GGML_TYPE_Q2_K - "type-1" 2-bit quantization in super-blocks containing 16 blocks, each block having 16 weight. Block scales and mins are quantized with 4 bits. This ends up effectively using 2.5625 bits per weight (bpw)
GGML_TYPE_Q3_K - "type-0" 3-bit quantization in super-blocks containing 16 blocks, each block having 16 weights. Scales are quantized with 6 bits. This end up using 3.4375 bpw.
GGML_TYPE_Q4_K - "type-1" 4-bit quantization in super-blocks containing 8 blocks, each block having 32 weights. Scales and mins are quantized with 6 bits. This ends up using 4.5 bpw.
GGML_TYPE_Q5_K - "type-1" 5-bit quantization. Same super-block structure as GGML_TYPE_Q4_K resulting in 5.5 bpw
GGML_TYPE_Q6_K - "type-0" 6-bit quantization. Super-blocks with 16 blocks, each block having 16 weights. Scales are quantized with 8 bits. This ends up using 6.5625 bpw
GGML_TYPE_Q8_K - "type-0" 8-bit quantization. Only used for quantizing intermediate results. The difference to the existing Q8_0 is that the block size is 256. All 2-6 bit dot products are implemented for this quantization type.
Refer to the Provided Files table below to see what files use which methods, and how.
Provided files
Name
Quant method
Bits
Size
Max RAM required
Use case
13B-HyperMantis.ggmlv3.q2_K.bin
q2_K
2
5.43 GB
7.93 GB
New k-quant method. Uses GGML_TYPE_Q4_K for the attention.vw and feed_forward.w2 tensors, GGML_TYPE_Q2_K for the other tensors.
13B-HyperMantis.ggmlv3.q3_K_L.bin
q3_K_L
3
6.87 GB
9.37 GB
New k-quant method. Uses GGML_TYPE_Q5_K for the attention.wv, attention.wo, and feed_forward.w2 tensors, else GGML_TYPE_Q3_K
13B-HyperMantis.ggmlv3.q3_K_M.bin
q3_K_M
3
6.25 GB
8.75 GB
New k-quant method. Uses GGML_TYPE_Q4_K for the attention.wv, attention.wo, and feed_forward.w2 tensors, else GGML_TYPE_Q3_K
13B-HyperMantis.ggmlv3.q3_K_S.bin
q3_K_S
3
5.59 GB
8.09 GB
New k-quant method. Uses GGML_TYPE_Q3_K for all tensors
13B-HyperMantis.ggmlv3.q4_0.bin
q4_0
4
7.32 GB
9.82 GB
Original llama.cpp quant method, 4-bit.
13B-HyperMantis.ggmlv3.q4_1.bin
q4_1
4
8.14 GB
10.64 GB
Original llama.cpp quant method, 4-bit. Higher accuracy than q4_0 but not as high as q5_0. However has quicker inference than q5 models.
13B-HyperMantis.ggmlv3.q4_K_M.bin
q4_K_M
4
7.82 GB
10.32 GB
New k-quant method. Uses GGML_TYPE_Q6_K for half of the attention.wv and feed_forward.w2 tensors, else GGML_TYPE_Q4_K
13B-HyperMantis.ggmlv3.q4_K_S.bin
q4_K_S
4
7.32 GB
9.82 GB
New k-quant method. Uses GGML_TYPE_Q4_K for all tensors
13B-HyperMantis.ggmlv3.q5_0.bin
q5_0
5
8.95 GB
11.45 GB
Original llama.cpp quant method, 5-bit. Higher accuracy, higher resource usage and slower inference.
13B-HyperMantis.ggmlv3.q5_1.bin
q5_1
5
9.76 GB
12.26 GB
Original llama.cpp quant method, 5-bit. Even higher accuracy, resource usage and slower inference.
13B-HyperMantis.ggmlv3.q5_K_M.bin
q5_K_M
5
9.21 GB
11.71 GB
New k-quant method. Uses GGML_TYPE_Q6_K for half of the attention.wv and feed_forward.w2 tensors, else GGML_TYPE_Q5_K
13B-HyperMantis.ggmlv3.q5_K_S.bin
q5_K_S
5
8.95 GB
11.45 GB
New k-quant method. Uses GGML_TYPE_Q5_K for all tensors
13B-HyperMantis.ggmlv3.q6_K.bin
q6_K
6
10.68 GB
13.18 GB
New k-quant method. Uses GGML_TYPE_Q8_K - 6-bit quantization - for all tensors
13B-HyperMantis.ggmlv3.q8_0.bin
q8_0
8
13.83 GB
16.33 GB
Original llama.cpp quant method, 8-bit. Almost indistinguishable from float16. High resource use and slow. Not recommended for most users.
Note: the above RAM figures assume no GPU offloading. If layers are offloaded to the GPU, this will reduce RAM usage and use VRAM instead.
How to run in llama.cpp
I use the following command line; adjust for your tastes and needs:
./main -t 10 -ngl 32 -m 13B-HyperMantis.ggmlv3.q5_0.bin --color -c 2048 --temp 0.7 --repeat_penalty 1.1 -n -1 -p "### Instruction: Write a story about llamas\n### Response:"
Change -t 10 to the number of physical CPU cores you have. For example if your system has 8 cores/16 threads, use -t 8.
Change -ngl 32 to the number of layers to offload to GPU. Remove it if you don't have GPU acceleration.
If you want to have a chat-style conversation, replace the -p <PROMPT> argument with -i -ins
I've had a lot of people ask if they can contribute. I enjoy providing models and helping people, and would love to be able to spend even more time doing it, as well as expanding into new projects like fine tuning/training.
If you're able and willing to contribute it will be most gratefully received and will help me to keep providing more models, and to start work on new AI projects.
Donaters will get priority support on any and all AI/LLM/model questions and requests, access to a private Discord room, plus other benefits.
Special thanks to: Luke from CarbonQuill, Aemon Algiz, Dmitriy Samsonov.
Patreon special mentions: Oscar Rangel, Eugene Pentland, Talal Aujan, Cory Kujawski, Luke, Asp the Wyvern, Ai Maven, Pyrater, Alps Aficionado, senxiiz, Willem Michiel, Junyu Yang, trip7s trip, Sebastain Graf, Joseph William Delisle, Lone Striker, Jonathan Leane, Johann-Peter Hartmann, David Flickinger, Spiking Neurons AB, Kevin Schuppel, Mano Prime, Dmitriy Samsonov, Sean Connelly, Nathan LeClaire, Alain Rossmann, Fen Risland, Derek Yates, Luke Pendergrass, Nikolai Manek, Khalefa Al-Ahmad, Artur Olbinski, John Detwiler, Ajan Kanaga, Imad Khwaja, Trenton Dambrowitz, Kalila, vamX, webtim, Illia Dulskyi.
Thank you to all my generous patrons and donaters!
(GGML and GPTQ are no longer in this repo and will be migrated to a separate repo for easier git download convenience)
Subjective testing shows quality results with KoboldAI (similar results are likely in Text Generation Webui, please disregard KAI-centric settings for that platform); Godlike preset with these tweaks - 2048 context, 800 Output Length, 1.3 Temp, 1.13 Repetition Penalty, AltTextGen:On, AltRepPen:Off, No Prompt Gen:On
Despite being primarily uncensored Vicuna models at its core, HyperMantis seems to respond best to the Alpaca instruct format. Speculatively due to manticore's eclectic instruct datasets generalizing the model's understanding of following instruct formats to some degree. What is known is HyperMantis responds best to the formality of Alpaca's format, whereas Human/Assistant appears to trigger vestigial traces of moralizing and servitude that aren't conducive for roleplay or freeform instructions.
Here is an example of what to place in KAI's Memory (or TGUI's equivalent) to leverage chat as a Roleplay Adventure.
[Define what the role of the named Human/AI are here, let's say our name is 'Player' and we named the AI 'Narrator']
Game Mode:Chat [Remember to name yourself and the AI and reference them in the instruction block]
### Instruction:
Make Narrator perform as a text based adventure game with Player as Narrator's user input. Make Narrator describe the scene, scenario, actions of characters, reactions of characters to the player's actions, and potential consequences of their actions and Player's actions when relevant with visually descriptive, detailed, and long storytelling. Allow characters and Player to converse to immerse Player in a rich narrative driven story. When Player encounters a new character, Narrator will name the new character and describe their behavior and appearance. Narrator will internally determine their underlying motivations and weave it into the story where possible.
### Response:
[Put A Carriage Return Here]
In KAI, this is why 'No Prompt Gen:On' is important; make your first entry a short writeup of your current situation, or simply reiterate Narrator is a text adventure game and Player is the input. Then your next entry, despite simply being a chat interface, it will kick off what will happen next for Narrator to riff off of. In TGUI, an equivalent setup works the same. Of course, tailor this to whatever you want it to be; instruct models can be as versatile as your imagination. If things go sideways have fun.
Possibly also useful as a regular chatbot, waifu, husbando, TavernAI character, freeform instruct shenanigans, it's whatever. 4bit-128g safetensor [Cuda] included for convenience, might do ggml. Mileage may vary, warranty void if the void stares back.
Credits:
manticore-13b [Epoch3] by openaccess-ai-collective
"Such as gravity's rainbow, sufficiently complex systems stir emergent behavior near imperceptible, uncanny; a Schrodinger's puzzlebox of what may be intrinsic or agentic. Best not to startle what black box phantoms there may be."
